[prev in list] [next in list] [prev in thread] [next in thread] 

List:       busybox
Subject:    Re: svn commit: trunk/busybox/init
From:       "JoSH Lehan" <krellan () gmail ! com>
Date:       2007-12-28 23:50:02
Message-ID: 242876bc0712281550p4c9d4b89jb1e6710fe7cd75b9 () mail ! gmail ! com
[Download RAW message or body]

On Dec 28, 2007 9:23 AM,  <busybox-request@busybox.net> wrote:
> Date: Fri, 28 Dec 2007 16:59:52 +0000
> From: Denys Vlasenko <vda.linux@googlemail.com>
> Subject: Re: svn commit: trunk/busybox/init
> To: busybox-cvs@busybox.net, busybox@busybox.net
> Message-ID: <200712281659.52839.vda.linux@googlemail.com>
> Content-Type: text/plain;  charset="iso-8859-15"
>
> I am catching such errors by running thousands of randomconfig builds
> before each release.

That's cool.

Does there already exist a utility to note the size of the resulting
busybox executable(s), along with the config file that was used, for
each randomconfig build?

If so, then over time, analysis of the results could eventually
determine the approximate cost, in bytes, for each of the config
options.  This would be useful information to know.

Of course, certain build options that have a huge influence on size
(static-ness, library-ness, and so on) would have to be locked down,
for comparisons to be meaningful.  For all of the other options, doing
repeated randomconfig builds would profile them over time.

It would start out very approximate, but then get better.  It would
basically take the size increase of the executable compared to a
baseline "allnoconfig" executable, and assign blame by dividing that
increase equally over each of the enabled config options.  Very crude,
but after iterating over those thousands of builds, eventually the
approximate size contribution for each option would be learned.

Josh
_______________________________________________
busybox mailing list
busybox@busybox.net
http://busybox.net/cgi-bin/mailman/listinfo/busybox
[prev in list] [next in list] [prev in thread] [next in thread]